|Publication number||US6308261 B1|
|Application number||US 09/016,692|
|Publication date||Oct 23, 2001|
|Filing date||Jan 30, 1998|
|Priority date||Jan 30, 1998|
|Also published as||DE69931288D1, DE69931288T2, EP0933698A2, EP0933698A3, EP0933698B1|
|Publication number||016692, 09016692, US 6308261 B1, US 6308261B1, US-B1-6308261, US6308261 B1, US6308261B1|
|Inventors||Dale C. Morris, Douglas B. Hunt|
|Original Assignee||Hewlett-Packard Company|
|Export Citation||BiBTeX, EndNote, RefMan|
|Patent Citations (12), Non-Patent Citations (1), Referenced by (49), Classifications (29), Legal Events (7)|
|External Links: USPTO, USPTO Assignment, Espacenet|
The present invention relates to computer instructions that are executed in computer systems. More specifically, the present invention relates to a computer instruction that determines whether accessing the contents of a register will cause execution of a program on a computer system to stall, thereby providing an indication of memory latency.
A typical computer system includes one or more processors, a memory system, and storage media. The memory system typically includes a main memory, and one or more cache memories coupled between the main memory and the processors. The storage media may comprise hard disk drives, floppy disk drives, CD-ROM drives, and the like.
A computer system performs a task by executing a computer program that is comprised of sequences of instructions. The performance of computer systems is greatly affected by the rate at which instructions are executed.
One common type of instruction is a load instruction that retrieves an operand from memory and stores the operand in a register of the processor. If the operand is stored in the cache memory, it will be retrieved quickly. This is known in the art as a “cache hit”. However, if the operand is only available in main memory, it may take considerably longer for the operand to be loaded into the register. This is known in the art as a “cache miss”. Many computer system will continue to execute instructions after a cache miss until encountering an instruction that requires the operand being loaded into the register from memory. Upon encountering such an instruction, the processor will stall execution of that instruction and wait for the load to finish. This technique is known in the art as “stall on use”, and cache memories that support this strategy are known as “non-blocking caches”.
The performance of processors is increasing faster than the performance of main memory systems. Current computer systems may require up to 100 processor cycles to retrieve an operand from main memory, and it is expected that future computer systems will require 1000 or more processor cycles. Accordingly, to achieve optimal system performance it is important to ensure that the processor has other instructions that can be executed while an operand is being retrieved from main memory. Ideally, the processor should never have to stall while waiting for a load to complete.
Some computer systems are capable of executing instructions out-of-order and are therefore able to defer stalling by executing other instructions while waiting for a load to complete. However, out-of-order execution requires complex hardware, and the window of instructions that are eligible to be executed out-of-order is relatively small (no more than a few hundred instructions). In addition, as more logic of a CPU is devoted to supporting out-of-order execution, less logic is available to perform the functions specified by the instructions.
Other techniques allow the processor to switch contexts when it is determined that the current stream of instructions cannot proceed because an operand is not available. For example, it is possible to maintain multiple contexts within the CPU, and simply switch to another context when a stall occurs. However, this approach requires extra hardware for maintaining multiple contexts, and like out-of-order execution, thereby reduces the hardware available to perform the functions specified by the instructions. It is also possible to generate an interrupt upon stalling and have software switch to some other thread or process. While this approach works well for longer latencies, such as retrieving data from a disk, the overhead incurred prohibits this approach when dealing with shorter latencies, such as accessing data from main memory.
Another technique is to model the latencies of various operations involved in executing instructions when a computer program is compiled. By scheduling instructions based on latency models, a compiler can significantly reduce stalling. For example, if a latency model indicates that a cache miss will require 30 CPU cycles, and a compiler can determine that a particular load instruction will generate a cache miss, a compiler can schedule a memory pre-fetch operation to retrieve the operand at least 30 cycles before the operand is needed.
Unfortunately, latencies are often difficult to predict at compile time. A compiler may not be able to determine whether a particular load will produce a cache miss or a cache hit. In addition, a single version of “shrink wrap” software is often marketed for a particular instruction set. However, such software may be executed on a wide variety of computer systems having substantial differences in performance and architecture. For example, a program written for the Intel x86 instruction set may be executed on computer systems ranging from a laptop computer having a PentiumŪ CPU, no external L2 cache, and fast page mode DRAM memory, to a computer workstation having multiple PentiumŪ II CPUs (with each CPU having an L1 and L2 cache) and synchronous DRAM memory.
Dynamic variable latencies were addressed by Mark Horowitz, Margaret Martonosi, Todd Mowry, and Michael Smith in a first paper entitled “Informing Loads: Enabling Software to Observe and React to Memory Behavior”, which was published in July of 1995, and in second paper entitled “Informing Memory Operations: Providing Memory Performance Feedback in Modern Processors. Both these papers are hereby incorporated by reference. These papers proposes a class of instructions called “informing load instructions”. An informing load instruction causes execution of the instruction immediately after the informing load instruction to be skipped if the load operation produces a cache hit, and executes the instruction immediately after the informing load instruction if the load operation produces a cache miss. Informing loads are non-blocking loads that do not stall execution. By scheduling a branch instruction immediately after an informing load operation, informing load instructions allow a compiler to schedule alternate action while a cache miss caused by an informing load operation is serviced.
While an informing load operation indicates whether a particular operand is in the first level cache, it does not provide an indication of the magnitude of the latency. The operand may be in an L2 cache, it may be in main memory, or it may be present as a dirty line in the cache of another CPU. Accordingly, informing load operations do not provide compilers with a method of scheduling alternate code threads based on the magnitude of the latency. Nor do informing loads provide a mechanism for a program to confirm that the contents of a register are available after the informing load itself has been executed.
The present invention is a computer system that includes a data structure that maintains availability status for registers of a processor of the computer system, wherein the availability status indicates whether an instruction attempting to read a particular register will stall. The computer system also includes instruction decode and execution circuitry that is capable of decoding and executing one or more instructions that alter a path of program execution based on the availability status of one or more of the registers.
In one embodiment, a latency probe instruction retrieves the availability status of a register from the data structure and stores the availability status in a register. Thereafter, a conditional branch instruction determines the path of program execution based on the availability status stored in the register. In another embodiment, a conditional branch instruction queries the data structure directly to determine the availability status of a register, and determines the execution path based on the availability status.
The present invention exposes the latency of memory operations to programs, thereby allowing a compiler to schedule alternate threads of execution based on varying latencies. By scheduling alternate threads of execution based on the availability of the contents of a register, stalling is minimized. If the compiler can accurately model the latencies caused by cache hits and misses, code can be scheduled to check for the availability of register contents at predefined intervals that correspond with predicted cache events. Alternatively, if the compiler cannot accurately model latencies, the compiler can schedule code to periodically check for the availability of register contents.
In addition, the present invention can also be used by a compiler to optimize code. The present invention provides a powerful method to collect data while a program is executing. The collected data can be used to optimize an execution file when a program is recompiled. The present invention can also be used by a compiler to schedule code that monitors performance, and in essence, recompiles itself “on-the-fly” as it executes.
One of the most compelling features of the present invention is that it is easy to implement in many modern computer architectures. Processors that support “stall on use” techniques and include “non-blocking caches” typically contain data structures capable of providing availability status for registers of the processor. In such a processor design, it is a relatively easy design task to encode an instruction that allows the availability status of registers to be tested by a conditional branch instruction.
FIG. 1 is a block diagram of a prior art computer system that includes a central processing unit (CPU), a level one (L1) cache memory unit, a level two (L2) cache memory unit, and a main memory unit.
FIG. 2 is a simplified illustration of a common prior art technique for determining whether a processor should suspend execution because the contents of a register are not available.
FIG. 3 is a simplified block diagram of CPU in accordance with the present invention.
FIG. 4 illustrates one simple application of a computer instruction provided by the present invention.
The present invention is a computer system capable of executing a computer instruction that determines whether an operand is available in a register. By executing the instruction of the present invention at various intervals, it is possible to probe the latency of a memory load operation.
FIG. 1 is a block diagram of a prior art computer system 10. Computer system 10 includes a central processing unit (CPU) 12, a level one (L1) cache memory unit 14, a level two (L2) cache memory unit 16, and a main memory unit 18. L1 cache memory unit 14 comprises a relatively small amount of fast, expensive memory that is closely coupled to CPU 12, and stores a subset of the memory contents of main memory unit 18. Many prior art CPUs include an L1 cache on the same integrated circuit as the CPU itself. Typically, a memory operand can be loaded from L1 cache memory unit 14 into a register of CPU 12 within one or two instruction clock cycles.
Compared with L1 cache memory unit 14, L2 cache memory unit 16 comprises a relatively larger (and relatively slower) amount of memory. L2 cache memory unit 16 also stores a subset of the contents of main memory unit 18. Typically, a memory operand can be loaded from L2 cache memory unit 16 within several instruction clock cycles.
Main memory unit 18 comprises a large amount of relatively slow memory. It may take 100 or more instructions clock cycles to load a memory operand from main memory unit 18 into a register of CPU 12.
While computer system 10 represents many prior art configurations, there are of course many other configurations. For example, it is common to have more or fewer cache memories. It is also common to have multiple processors, with processors possibly sharing some caches and having exclusive access to other caches.
Note that the latency of a memory operation varies depending upon whether a particular memory operand is stored in a particular cache memory unit. Often, a compiler will be able to predict if an operand is in a particular cache memory unit, and schedule code accordingly. However, it is also common that a compiler will not be able to determine whether an operand will be in any of the cache memories.
In many prior art computer systems, the CPU does not stall when a load instructions that loads an operand from memory into a register produces a cache miss. Rather, the CPU continues to execute instructions until reaching an later instruction that requires the contents of that register. If the operand has been loaded from memory, the later instruction is executed. If the operand has not been loaded from memory, execution (of at least the later instruction) is suspended until the cache miss is serviced, at which point execution resumes with the later instruction. This technique is known in the art as “stall on use”, and cache memories that support this strategy are known as “non-blocking caches”.
FIG. 2 is a simplified illustration of a common prior art technique for determining whether a processor should suspend execution because the contents of a register are not available. In FIG. 2, CPU 12 includes registers 20, unavailable flags 22, and stall signal unit 24. Other CPU functions, such as those provided by instruction decode units, floating point execution units, integer execution units, memory access units, and the like are represented generically as box 26.
To understand how CPU 12 determines whether execution should be suspended, assume that CPU 12 begins to execute a load instruction that loads the contents of the memory location specified by the contents of register 1 (r1) into register 0 (r0). Such an instruction can be represented as:
Further assume that the contents of the memory location specified by the contents of r1 are in L1 cache memory unit 14. When instruction is executed, the unavailable flag 22 associated with r0 is set to “1” to indicate that the contents of r0 are unavailable. The next instruction clock cycle, the r0 will be loaded from L1 cache memory unit 14 and the unavailable flag 22 associated with r0 will be cleared to “0” to indicate that the contents of r0 are available.
Now assume that the next instruction is:
add r3=r0, r2
This instruction will add the contents of r0 and r2 and store the result in r3. Assume that the contents of r2 are available, and therefore the unavailable flag 22 associated with r2 is cleared to “0”. When the instruction begins to execute, stall signal unit 24 generates a stall signal based on the unavailable flags 22 associated with r0 and r2. Since the contents of both registers are available, the stall signal is not asserted and the add instruction is executed.
Now assume that the same sequence of instructions is executed, except that the contents of the memory location specified by the contents of r1 are stored in main memory unit 18, but not cache memory units 14 or 16. Accordingly, the contents of r0 will not be available at the next instruction clock cycle when the add instruction is executed. Stall signal unit 24 will assert a stall signal, which will be used by CPU 12 to stall execution until the memory contents are loaded from main memory unit 18 into r0. When the memory contents are loaded, execution of the add instruction will be completed.
Of course, the example discussed above with reference to FIG. 2 is simplified. In a CPU supporting pipelined execution, the function provided by unavailable flags 22 is provided by a more complex mechanism. Consider a CPU having a five-stage pipeline comprised of the stages of instruction fetch (IF), instruction decode and register fetch (RD), execution (EX), data memory access (ME), and register write back (WB). Assume that such a processor also has a table of outstanding memory read transactions, with each active entry in the table including a register address specifying the destination register that will receive the results of the transaction. If an instruction is going to stall because the contents of a register are not available, it will stall at the RD stage. To generate a stall signal at the RD stage, it must be determined if any of the instructions in later stages of the pipeline or the transactions in the table of outstanding memory read transactions will store data in the register that is read at the RD stage. Accordingly, in such a processor the function performed by unavailable flags 22 and stall signal unit 24 in FIG. 2 is performed by a series of comparators and OR gates that compare the source register of the instruction at the RD stage with the destination registers of instructions at later stages and the destination registers stored in active entries of the table of outstanding memory read transactions, with the stall signal being generated by OR'ing together the results from all the comparators. Of course, those knowledgeable in the art will recognize that a variety of different data structures and techniques may by used to keep track of whether the contents of a register will be available. In addition, those knowledgeable in the art will also recognize that execution need only be stalled for the instruction that requires the register contents that are unavailable. Execution of other instructions, as well as other unrelated CPU functions, may continue.
The present invention provides a mechanism for allowing a computer program to ascertain whether the contents of a register are available, or whether accessing that register will cause execution to stall. FIG. 3 is a simplified block diagram of CPU 28. FIG. 3 illustrates how CPU 12 of FIG. 1 may be modified to implement the present invention. Similar to CPU 12, CPU 28 includes registers 30, unavailable flags 32, and stall signal unit 34. CPU 28 also includes multiplexer 36, which is capable of routing the contents of any of the unavailable flags 32 into any of the registers 30. Finally, the CPU functions represented by box 38 include circuitry capable of decoding an instruction in accordance with the present invention (discussed below) and controlling multiplexer 36.
Multiplexer 36 represents any mechanism capable of linking an unavailable flag into a register. For understanding the present invention, separate data paths are shown linking multiplexer 35 with unavailable flags 32 and register 30. However, those skilled in the are will recognize that the functions represented by multiplexer 36 would likely be implemented by driving a bit of a common data path leading into the registers to a value stored in one of the unavailable flags. In addition, in a pipelined processor, the function performed by unavailable flags 32 may be performed by comparators that compare a source register address with destination register addresses in later stages of the pipeline and in a table of outstanding memory read transactions, as discussed above with reference to FIG. 2. As used herein, the term “unavailable flag” includes availability status generated in such a manner.
The circuitry required to decode a computer instruction in accordance with the present invention is represented by box 38. Those skilled in the prior art will recognize how to adapt the teachings herein to properly define an operation code for an instruction in accordance with the present invention, and design appropriate decode logic to decode the instruction and operate multiplexer 36 to implement the functionality described herein.
Consider an instruction in accordance with the present invention called a latency probe instruction. One format for such an instruction is:
When an iprobe instruction is executed, the unavailable flag 32 associated with rj is stored in ri. Accordingly, the contents of ri will be “0” if the contents of rj are available, and the contents of ri will be “1” if the contents of rj are unavailable. The contents of ri may then be tested via a conditional branch instruction to determine whether to execute an instruction that accesses register rj. Of course, other instructions in accordance with the present invention are also possible. For example, in one embodiment the processor may include condition (or predicate) registers that are used to control branching behavior, and an iprobe instruction may be defined to transfer the specified unavailable flag 32 to a specified condition register. In another embodiment, a class of branch instructions may be defined to treat the unavailable flags as registers upon which a conditional branch may be based. One format for such an instruction is:
br rj.uf, miss
Such an instruction would branch to the address miss only if the unavailable flag associated with rj were unavailable.
Of course, many other instruction encodings are possible. The key to the present invention is that the data structure or technique that is used to determine whether the contents of a register are available may be queried by a computer instruction to determine whether an instruction that accesses the register will stall, and the thread of execution can be varied based on the results of the query. Accordingly, the present invention exposes the latency of register load operations to a compiler, thereby allowing a compiler to schedule alternate threads based on whether the contents of a register will be available.
FIG. 4 illustrates one simple application of the computer instruction of the present invention. Assume that a CPU in accordance with the present invention, such as CPU 28 in FIG. 3, is present in a computer system having L1 and L2 cache units, and a main memory unit, such as computer system 10 in FIG. 1. Further assume that a compiler has accurate latency models that predict the latencies of an L1 and L2 cache hits and misses, and the compiler must schedule code which loads the memory contents specified by the contents of ri into r0, and adds the contents of r0 and r2 and stores the results into r3. Also assume that the compiler is able to predict that the contents of r2 will be available.
First, the compiler schedules code segment 40. The first instruction in code segment 40 loads the memory contents specified by the contents of r1 into r0. The next instruction is an iprobe instruction that loads register r4 with the contents of the unavailable flag 32 associated with r0. The next instruction is a branch instruction that tests the contents of register r4. If the contents of r0 are available, r4 will be “0” and the branch will fail, which is consistent with an L1 cache hit. The next instruction adds the contents of r0 and r2 and stores the results into r3. The remainder of the code in segment 40 is code that has been optimized based on the assumption of an L1 cache hit. If the contents of r0 are unavailable, r4 will be “1” and the branch will direct execution to code segment 42, which is at address miss—1.
Code segment 42 first executes a series of instructions that are optimized based on the assumption of an L1 cache miss. The compiler assumes that the memory contents being loaded into r0 are stored in the L2 cache and identifies and schedules other instructions that may be executed during that time the compiler expects the memory contents specified by the contents of r1 are possibly retrieved from the L2 cache. After these instructions have been executed, another iprobe instruction is executed that loads r4 with the unavailable flag 32 associated with r0. The next instruction is a branch instruction which tests the contents of r4. If the contents of r0 are available, the branch will fail, which is consistent with an L2cache hit. Execution will then continue with add instruction and other instructions that were scheduled based on an L2 cache hit. However, if the contents of r0 are not available, the branch will be taken to code segment 44, which is at address miss—2.
In code segment 44, the compiler has scheduled a first portion of code that is optimized for an L2 cache miss. The compiler's latency model predicts that after the first portion has been executed, sufficient time will have passed to allow the L2 cache miss to be serviced, and the contents of r0 will be available. Thereafter, the add instruction is executed along with other instructions that the compiler scheduled to be executed based on an assumption of an L2 cache miss.
The above example illustrates how a compiler can use the instruction of the present invention to schedule code around a register load having an unknown, but limited and defined set of possible latencies. By scheduling lprobe instructions at periodic intervals, it is also possible to schedule code when the set of latencies cannot be defined, such as when data being loaded is being held in a dirty cache line of another CPU.
In addition, the present invention can also be used by a compiler to optimize code. For example, it is known in the art to collect data during a profile-based optimization (PBO) execution session. The program is then recompiled to using the collected data to produce an optimized executable version of the program. The present invention provides a powerful method to collect data during the PBO session, since lprobe instruction may be used to detect cache misses, and measure memory latencies.
Another application for the present invention is continuous run-time optimization. It is anticipated that the number of computer system configurations will continue to proliferate. In the future, it may be common for a single compiled computer program to be expected to execute efficiently on various computer systems have different memory configurations and numbers of CPUs. It may not be practical to anticipate all possible configurations when a program is compiled. Therefore, it is expected that programs will include code to monitor performance, and in essence recompile “on-the-fly” as they execute. The present invention provides a mechanism for a program to monitor its performance, and reschedule its code accordingly. Unlike prior techniques, the present invention not only allows a program to determine if a register load produces cache hit, it also allows a program to measure how long it takes for the content of the register to become available.
Besides being able to use the present invention to change the path of execution of a program when the contents of a register are not available, the present invention can also be used to control cache prefetch operations in a computer system capable of executing instructions out-of-order. For example, assume that a branch condition is determined based on a value loaded from memory. If the loaded value is in the cache, it will not be long before the processor computes the result and resolves the branch condition. Accordingly, the potential benefit of starting a speculative cache prefetch operation is small since it will soon be known whether a the load instruction requiring the prefetch is on the execution path. Since the benefit is small, it may be desirable to avoid the prefetch operation to avoid polluting the cache with unneeded data. On the other hand, if the load required to determine the branch condition produces a cache miss, it will take some time for the processor to resolve the branch condition. In this situation, the benefit of performing a speculative prefetch operation for a load instruction along a predicted (but not yet confirmed) execution path is greater. The present invention provides a low-overhead mechanism to allow a compiler to schedule code that determines whether a branch condition will be resolved quickly, and thereby decide whether to follow an execution path that includes speculative prefetch operations.
Perhaps one of the greatest benefits of the present invention is that it is relatively simple to implement and requires little additional circuitry in processor architectures that support “stall on use” techniques and include “non-blocking” caches. Such processors already include substantial circuitry to generate a stall signal. In accordance with the present invention, all that must be added to such a processor is an instruction that activates that circuitry to produce a stall signal for a particular register, and stores the result of the stall signal in another register (or branches based on the stall signal). Accordingly, the present invention can be implemented by a handful of gates that decode the instruction and divert the already present stall signal into the data path of a destination register.
Although the present invention has been described with reference to preferred embodiments, workers skilled in the art will recognize that changes may be made in form and detail without departing from the spirit and scope of the invention.
|Cited Patent||Filing date||Publication date||Applicant||Title|
|US5142631 *||Feb 3, 1989||Aug 25, 1992||Digital Equipment Corporation||System for queuing individual read or write mask and generating respective composite mask for controlling access to general purpose register|
|US5251306||Jan 16, 1990||Oct 5, 1993||Advanced Micro Devices, Inc.||Apparatus for controlling execution of a program in a computing device|
|US5361337 *||May 8, 1992||Nov 1, 1994||Sun Microsystems, Inc.||Method and apparatus for rapidly switching processes in a computer system|
|US5530825 *||Apr 15, 1994||Jun 25, 1996||Motorola, Inc.||Data processor with branch target address cache and method of operation|
|US5802386 *||Nov 19, 1996||Sep 1, 1998||International Business Machines Corporation||Latency-based scheduling of instructions in a superscalar processor|
|US5838986 *||Sep 25, 1997||Nov 17, 1998||Seiko Epson Corporation||RISC microprocessor architecture implementing multiple typed register sets|
|US5887166 *||Dec 16, 1996||Mar 23, 1999||International Business Machines Corporation||Method and system for constructing a program including a navigation instruction|
|US5905889 *||Mar 20, 1997||May 18, 1999||International Business Machines Corporation||Resource management system using next available integer from an integer pool and returning the integer thereto as the next available integer upon completion of use|
|US5913059 *||Jul 25, 1997||Jun 15, 1999||Nec Corporation||Multi-processor system for inheriting contents of register from parent thread to child thread|
|US5933627 *||Jul 1, 1996||Aug 3, 1999||Sun Microsystems||Thread switch on blocked load or store using instruction thread field|
|US6016542 *||Dec 31, 1997||Jan 18, 2000||Intel Corporation||Detecting long latency pipeline stalls for thread switching|
|EP0747816A2||May 31, 1996||Dec 11, 1996||International Business Machines Corporation||Method and system for high performance multithread operation in a data processing system|
|1||Gregory T. Byrd, et al., Multithreaded Processor Architectures, Aug. 1, 1995, IEEE Spectrum, IEEE Inc., vol. 32, No. 8, pp. 38-46.|
|Citing Patent||Filing date||Publication date||Applicant||Title|
|US6789100||Feb 8, 2002||Sep 7, 2004||Mips Technologies, Inc.||Interstream control and communications for multi-streaming digital processors|
|US6973561 *||Dec 4, 2000||Dec 6, 2005||Lsi Logic Corporation||Processor pipeline stall based on data register status|
|US7020879||May 14, 1999||Mar 28, 2006||Mips Technologies, Inc.||Interrupt and exception handling for multi-streaming digital processors|
|US7035997||Jul 14, 2000||Apr 25, 2006||Mips Technologies, Inc.||Methods and apparatus for improving fetching and dispatch of instructions in multithreaded processors|
|US7100157 *||Sep 24, 2002||Aug 29, 2006||Intel Corporation||Methods and apparatus to avoid dynamic micro-architectural penalties in an in-order processor|
|US7134139 *||Feb 12, 2002||Nov 7, 2006||International Business Machines Corporation||System and method for authenticating block level cache access on network|
|US7137111 *||Nov 28, 2001||Nov 14, 2006||Sun Microsystems, Inc.||Aggressive prefetch of address chains|
|US7206922||Dec 30, 2003||Apr 17, 2007||Cisco Systems, Inc.||Instruction memory hierarchy for an embedded processor|
|US7237093 *||Jun 16, 2000||Jun 26, 2007||Mips Technologies, Inc.||Instruction fetching system in a multithreaded processor utilizing cache miss predictions to fetch instructions from multiple hardware streams|
|US7257814||Jun 12, 2000||Aug 14, 2007||Mips Technologies, Inc.||Method and apparatus for implementing atomicity of memory operations in dynamic multi-streaming processors|
|US7360064||Dec 10, 2003||Apr 15, 2008||Cisco Technology, Inc.||Thread interleaving in a multithreaded embedded processor|
|US7441101||Feb 5, 2004||Oct 21, 2008||Cisco Technology, Inc.||Thread-aware instruction fetching in a multithreaded embedded processor|
|US7454747 *||Feb 7, 2003||Nov 18, 2008||Sun Microsystems, Inc.||Determining maximum acceptable scheduling load latency using hierarchical search|
|US7467385||Mar 21, 2006||Dec 16, 2008||Mips Technologies, Inc.||Interrupt and exception handling for multi-streaming digital processors|
|US7516306||Oct 5, 2004||Apr 7, 2009||International Business Machines Corporation||Computer program instruction architecture, system and process using partial ordering for adaptive response to memory latencies|
|US7529907||Oct 22, 2007||May 5, 2009||Mips Technologies, Inc.||Method and apparatus for improved computer load and store operations|
|US7650605||Feb 20, 2007||Jan 19, 2010||Mips Technologies, Inc.||Method and apparatus for implementing atomicity of memory operations in dynamic multi-streaming processors|
|US7707391||Apr 6, 2006||Apr 27, 2010||Mips Technologies, Inc.||Methods and apparatus for improving fetching and dispatch of instructions in multithreaded processors|
|US7707554||Jun 30, 2004||Apr 27, 2010||Oracle America, Inc.||Associating data source information with runtime events|
|US7711936 *||Aug 28, 2007||May 4, 2010||Sun Microsystems, Inc.||Branch predictor for branches with asymmetric penalties|
|US7735073||May 6, 2004||Jun 8, 2010||Oracle International Corporation||Method and apparatus for data object profiling|
|US7765546||Aug 18, 2004||Jul 27, 2010||Mips Technologies, Inc.||Interstream control and communications for multi-streaming digital processors|
|US7827543||May 6, 2004||Nov 2, 2010||Oracle America, Inc.||Method and apparatus for profiling data addresses|
|US7900207||Nov 19, 2008||Mar 1, 2011||Mips Technologies, Inc.||Interrupt and exception handling for multi-streaming digital processors|
|US7926062||Apr 29, 2009||Apr 12, 2011||Mips Technologies, Inc.||Interrupt and exception handling for multi-streaming digital processors|
|US8065665||May 6, 2004||Nov 22, 2011||Oracle America, Inc.||Method and apparatus for correlating profile data|
|US8122439||Aug 9, 2007||Feb 21, 2012||International Business Machines Corporation||Method and computer program product for dynamically and precisely discovering deliquent memory operations|
|US8468540||Mar 7, 2011||Jun 18, 2013||Bridge Crossing, Llc||Interrupt and exception handling for multi-streaming digital processors|
|US9003169||Mar 16, 2009||Apr 7, 2015||International Business Machines Corporation||Systems and methods for indirect register access using status-checking and status-setting instructions|
|US20020062385 *||Aug 22, 2001||May 23, 2002||Dowling Eric Morgan||Negotiated wireless peripheral systems|
|US20030101336 *||Jan 16, 2002||May 29, 2003||Sun Microsystems, Inc.||Technique for associating instructions with execution events|
|US20030105942 *||Nov 28, 2001||Jun 5, 2003||Sun Microsystems, Inc.||Aggressive prefetch of address chains|
|US20030154412 *||Feb 12, 2002||Aug 14, 2003||International Business Machines Corporation||System and method for authenticating block level cache access on network|
|US20040060040 *||Sep 24, 2002||Mar 25, 2004||Collard Jean-Francois C.||Methods and apparatus to avoid dynamic micro-architectural penalties in an in-order processor|
|US20040158826 *||Feb 7, 2003||Aug 12, 2004||Sun Microsystems, Inc.||Determining maximum acceptable scheduling load latency using hierarchical search|
|US20040226011 *||May 8, 2003||Nov 11, 2004||International Business Machines Corporation||Multi-threaded microprocessor with queue flushing|
|US20050081214 *||Aug 18, 2004||Apr 14, 2005||Nemirovsky Mario D.||Interstream control and communications for multi-streaming digital processors|
|US20060101249 *||Oct 5, 2004||May 11, 2006||Ibm Corporation||Arrangements for adaptive response to latencies|
|US20060130012 *||Nov 9, 2005||Jun 15, 2006||Matsushita Electric Industrial Co., Ltd.||Program conversion device, program conversion and execution device, program conversion method, and program conversion and execution method|
|US20070061619 *||Mar 21, 2006||Mar 15, 2007||Nemirovsky Mario D||Interrupt and exception handling for multi-streaming digital processors|
|US20070260856 *||May 5, 2006||Nov 8, 2007||Tran Thang M||Methods and apparatus to detect data dependencies in an instruction pipeline|
|US20070294702 *||Feb 20, 2007||Dec 20, 2007||Mips Technologies, Inc.||Method and apparatus for implementing atomicity of memory operations in dynamic multi-streaming processors|
|US20090044176 *||Aug 9, 2007||Feb 12, 2009||International Business Machine Corporation||Method and Computer Program Product for Dynamically and Precisely Discovering Deliquent Memory Operations|
|US20090063831 *||Aug 28, 2007||Mar 5, 2009||Sun Microsystems, Inc.||Branch predictor for branches with asymmetric penalties|
|US20090125660 *||Nov 19, 2008||May 14, 2009||Mips Technologies, Inc.||Interrupt and Exception Handling for Multi-Streaming Digital Processors|
|US20090235035 *||Mar 16, 2009||Sep 17, 2009||International Business Machines Corporation||Computer program instruction architecture, system and process using partial ordering for adaptive response to memory latencies|
|US20090241119 *||Apr 29, 2009||Sep 24, 2009||Nemirovsky Mario D||Interrupt and Exception Handling for Multi-Streaming Digital Processors|
|US20110154347 *||Jun 23, 2011||Nemirovsky Mario D||Interrupt and Exception Handling for Multi-Streaming Digital Processors|
|CN100407137C||Nov 14, 2005||Jul 30, 2008||国际商业机器公司||推测性地执行程序指令的方法、微处理器及系统|
|U.S. Classification||712/219, 712/E09.053, 712/234, 718/108, 712/E09.049, 712/E09.047, 712/228, 712/E09.032, 712/E09.046, 712/217|
|International Classification||G06F9/30, G06F12/08, G06F9/45, G06F9/38, G06F9/32, G06F9/308|
|Cooperative Classification||G06F9/30047, G06F9/3836, G06F9/3851, G06F9/3857, G06F9/3824, G06F9/383, G06F9/3838|
|European Classification||G06F9/30A2P, G06F9/38E1, G06F9/38E, G06F9/38D, G06F9/38E4, G06F9/38D2|
|Jun 4, 1998||AS||Assignment|
Owner name: HEWLETT-PACKARD COMPANY, CALIFORNIA
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:MORRIS, DALE C.;HUNT, DOUGLAS B.;REEL/FRAME:009252/0744;SIGNING DATES FROM 19980327 TO 19980330
|Jan 16, 2001||AS||Assignment|
|Apr 25, 2005||FPAY||Fee payment|
Year of fee payment: 4
|Apr 23, 2009||FPAY||Fee payment|
Year of fee payment: 8
|Sep 22, 2011||AS||Assignment|
Owner name: HEWLETT-PACKARD DEVELOPMENT COMPANY, L.P., TEXAS
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:HEWLETT-PACKARD COMPANY;REEL/FRAME:026945/0699
Effective date: 20030131
|Jan 13, 2012||AS||Assignment|
Owner name: HTC CORPORATION, TAIWAN
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:HEWLETT-PACKARD DEVELOPMENT COMPANY, L.P.;REEL/FRAME:027531/0218
Effective date: 20111213
|Mar 14, 2013||FPAY||Fee payment|
Year of fee payment: 12